ChatGPT: a force for good or evil?

Friday, 17 February 2023

ChatGPT may not be perfect, but it's proving a faster learner.

EXPLAINED: Artificial intelligence tool ChatGPT promises to help automate the creation of letters, marketing copy, computer programs, university assignments and even music and art.

The way it works is simple. Visit the ChatGPT website and type in a question or ask it to churn out something, and off it goes.

But while it is proving a useful and fun tool for many – especially while it’s available for free – it has also given anyone who regurgitates, mashes up or seeks out information for a living pause for a number of dark thoughts.

The rub is that it has only just started “learning” and it already seems unnervingly able.

* I’m trying to make myself feel better after my interview with a robot

* What is ChatGPT and should you be using it for marketing?

The centuries-old question of whether machines are here to assist or replace us at work is bubbling up again, as are some more existential concerns, including whether ChatGPT could herald a surge in misinformation and disinformation.

Could ChatGPT be used to disseminate misinformation?

Yes, like any technology, ChatGPT could potentially be used to disseminate misinformation.

ChatGPT is a language model that generates text based on the patterns it learned from the vast amounts of data it was trained on. If the inputs provided to ChatGPT are biased or inaccurate, the responses it generates may also be biased or inaccurate.

ChatGPT has provided a wake-up call over the potential of artificial intelligence.

In addition, if ChatGPT is programmed or trained to generate text with a specific agenda or to promote certain beliefs, it could be used to intentionally spread misinformation or propaganda.

It is important to note that ChatGPT is not inherently biased or malicious.

Rather, it is the responsibility of the individuals and organisations that use and train the model to ensure that the inputs and outputs are accurate and unbiased.

In order to combat the spread of misinformation, it is important to critically evaluate the sources of information and fact-check claims before accepting them as true.

Hmm. Who wrote that?

OK, you guessed.

Not me. That was the disarmingly honest response that ChatGPT gave when I asked it whether ChatGPT could be used to disseminate misinformation.

If the sophistication of the reply came as any surprise, welcome to the world of “generative artificial intelligence”.

Artificial intelligence (AI) has been on an onward march over the decades as the power of computers has relentlessly increased and the skills of scientists in exploiting that power to address new challenges have grown.

But it is probably fair to say AI has been leaping new hurdles at an increasingly faster pace.

In 1951, Oxford University academicChristopher Strachey taught a computer to play draughts.

By 1997, IBM’s Deep Blue supercomputer, which cost an estimated $100 million to develop, was able to beat grandmaster Garry Kasparov at chess.

Today, no human stands a chance of beating the higher levels of chess-playing program Stockfish on a smartphone, and the frontiers of AI and machine-learning have moved on to using computers to address more subtle challenges.

Such as?

All sorts of useful and sometimes controversial things.

Making complex medical diagnoses such as detecting cancers from mammograms, identifying individuals from CCTV footage, and predicting the risk of someone defaulting on a loan to name but a few.

But probably the biggest frontier in AI lies in using computers to convincingly mimic human intelligence by manipulating language and speech.

And ‘generative AI’?

It describes the use of AI to generate what is these days commonly called “content”.

Examples include the feature on gmail that suggests a quick response to incoming emails and Google Translate, which uses AI to translate speech and text, as well as productivity tools such as Otter.ai that are designed to automatically generate transcripts from recorded speech.

ChatGPT is algorithm that has been trained on a vast amount of information that is freely available online to answer questions or perhaps write your university essay, marketing copy or a note to your boss asking for a raise.

It does that by using deep-learning algorithms, neural networks, and natural language processing to analyse patterns in texts, essentially to work out what words might come next.

ChatGPT won’t usually give the exact same answers or generate the same responses twice though.

It assesses how it should respond based on the probability of its responses being what might be wanted, but includes an element of “randomisation” to keep them varied.

What are the concerns about misinformation?

One is that ChatGPT, or engines like it, could be used to “weaponise” misinformation by churning out one-sided arguments.

Asked what the drawbacks of wind power were, for example, ChatGPT listed their visual impact, noise pollution, the risks to bird life and bats, their initial upfront investment, and so on.

Companies such as Google and Microsoft are making big investments in generative AI.

While it contextualised its response by noting wind power was a promising, renewable and clean energy source, those balancing comments could be easily stripped out from its response before being disseminated.

What else?

Companies such as Google and Microsoft are making big investments in generative AI in part because it could be used to change the experience provided by internet search engines.

Currently, if sometimes types a query into a search engine such as Google they will be provided with pages of links relevant to the question or term they have keyed in.

To people born before the internet that’s still something close to magic, but in the future that could look rather clunky and outmoded.

Search engines that put generative AI front and centre could instead be designed to return a single unified response rather than just dump a list of raw links.

Arguably that could reduce the chances of people being confronted with misinformation or disinformation online, but it also creates a risk that people could be exposed to information that is less diverse.

Moreover, ChatGPT notes it could learn from a user's search history to provide “customised recommendations based on their preferences and interests”.

“This can lead to a more personalised and engaging search experience,” it states.

The flip side is that the responses provided by generative search engines could reinforce people’s preferences and prejudices leading them deeper down rabbit holes.

Should we be worried?

On guard, at least.

It has always been possible for people to choose to mainly source information that will reinforce their outlook, whether that be by only tuning in to a particular news channel or a favoured flavour of talkback radio, clicking on personalised links dished up on Facebook, or simply through the company they choose to keep.

And as ChatGPT will tell you – with or without generative AI – people often find it easier to believe what they want to believe.

ChatGPT has some controls to guard against misuse. It refuses to make up a note to send to your boss if you want to throw a sickie for example, but it will set out the arguments that Russia was justified in invading Ukraine, albeit with the disclaimer that those arguments are widely criticised.

Looking on the bright side, generative AI could speed up the dissemination of balanced and comprehensive information that is easy to understand.

The nagging concern may be the sheer volume of high-quality misinformation tools such as ChatGPT could cheaply pump out, if instructed – especially if it became the default delivery mechanism for internet search.